Tiny Desk AI: how to make the tiny desk concert video

Tiny Desk AI takes your photo and puts you at a tiny desk: a small desk in a cramped corner, a house band, and a short performance already filmed. Your photo identifies the face in one seat — the whole performance comes from the prepared clip, so the camera moves and the timing stay on the beat.

You only need one photo. The lead singer is the only required seat; the other four are optional, and a seat you leave empty keeps the performer who is already in the clip.

What the tiny desk format is

The Tiny Desk Concert is NPR's video series: artists play a short set at a small desk in a cramped NPR office corner, backed by the house band, with posters and records on the walls behind them. The format started in 2008 and became a weekly video franchise with NPR Music in 2014, and its 2017 Despacito performance was the most-watched video on YouTube at the time.

The AI version that grew out of it is the short one — a few seconds of that room, with your face in it. Tiny Desk AI is an independent fan-made parody: it is not affiliated with, endorsed by, or sponsored by NPR, and every clip it produces is AI-generated.

The five seats

The prepared clip has five performers in it. Each seat takes one photo, and the model replaces that performer's face with yours while keeping the body, the clothes and the instrument from the clip.

A seat you leave empty is not filled with anybody new. The performer already in the clip stays exactly as they are — same face, same clothes, same place in the shot.

How to make a Tiny Desk AI video

1. 1Open the generator and upload one clear photo for the lead singer.
2. 2Optionally add photos for any other seat you want to change.
3. 3Write what anyone should wear or look like, if you want to change it.
4. 4Pay once. The render takes five to eight minutes.
5. 5Download the vertical file and post it with the AI label on.

The photo that works

One adult, facing the camera, face sharp and reasonably large in frame, with a plain background behind them. The face is the part that has to survive fifteen seconds of movement — a small or blurred face gets reinvented by the model, and the reinvention drifts from frame to frame.

• One person only. Two people in frame get averaged into a single shape.
• Face clear and lit. Eyes visible if you can manage it.
• Sharp, not mid-movement. Motion blur is the most common cause of a soft head.
• Plain background. Clutter in the source carries into the render.

What the photo cannot fix

Two things the generator deliberately leaves alone, because the prepared clip owns them: the performance and the camera. Your photo is the face, not the act. If you want different playing, no photo will get you there.